Robots do not fail very often, except when the motors used are too small; the joints use the energy wastefully. A hinge that is simple to model in software may nevertheless be wrong as a task requiring larger actuators, smaller tolerances, and always being correct. Designed to support the weight of a knee-inspired joint in geometry, a joint architecture at Harvard has rolling contact surfaces that are shaped to guide motion and focus forces where they are needed.
This mechanism begins with a type of joint called rolling contact joints, in which paired curved surfaces roll and slide together and are held together by flexible connectors. This type of rolling-sliding combination is the silent cause of the knee being able to support, change instantaneous center of rotation, and still feel smooth, in human anatomy. The concept has been present in robotics, primarily in the form of circular rolling contacts, which have the benefit of being practical, but restricted by the fact that a circle is able to capture only one simple correlation between angular position, trajectory, and the force delivered.
The Harvard approach substitutes the process of “choose a joint, then control around its flaws” its defects with an optimization program that develops the joint to be used in a given task. The design is aimed at two concurrent design surfaces, namely noncircular rolling forms and internal pulley design that defines a transmission behaviour. The authors of the paper Noncircular rolling contact joints enable programmed behavior in robotic linkages explain the possibility to program desired end-effector paths and variable mechanical advantage into the contact geometry without violating realistic constraints like size limits and convexity required to be manufactured.
“Whenever you have some robot, and you have an idea of what it needs to do – maybe it’s a walking robot – you can start to think about the best places to output force,” said Colter Decker, first author of the study. “If we can embed those decisions into the mechanics of the robot itself, then we can create robots that are more efficient. They can use smaller actuators because the energy is targeted specifically where it needs to be.”
That “targeting” would appear in the alignment with conventional assistive hinges usually opposing the body. A human knee is not rotating around a rigid axis; the point on it is moving along a path of rolling cartilage surfaces and sliding. The team showed a 99.6percent reduction in error of alignment compared with revolute joints, and a 99.3percent reduction compared with circular rolling-contact designs by optimizing a knee-like rolling contact joint to follow the same path.
The identical programmable mechanics are converted to grip strength with no increment in actuator input. The two-finger gripper prototype developed by the researchers based on the focus on optimized rolling contacts allowed the researchers to gain more than 3.5x the load-carrying capacity compared to a similar gripper with a circular joint. It is not merely that I have “stronger fingers,” but that through the modification of the transmission of force across the stroke, to suit the geometry of the grasp, I have had to induce the controller to adapt to it.
Robert J. Wood defined that more general purpose: the division of labor between mechanics and software: “We aim to offload as much motion control as possible to the mechanics and materials of the robot, so that the control system can focus on task-level goals. Colter’s methods do exactly that, and in a very elegant way, both mathematically and mechanically.”
The peculiarity of the approach that makes it unusual is that the joint is being designed as an interface of the design and not a preset catalog part. When a profile of forces and a profile of trajectory are inputted, walking push-off, a gentle wrap-and-squeeze grasp, or a prosthetic-like motion, then the optimizer finds the rolling surfaces and pulley shapes that replicate these, with no need to specify an explosion in sensors, calibration procedures or controller sophistication.